Oh it's a very very good question, I'm glad that someone asked it. Because there is a completely mind bending paradox about trends, that I believe no one has yet realized, as I never saw it discussed anywhere. I've spotted it in IT, with trends about programming languages, on Google Trends for example.

Let's say that you want to compare 2 programming languages: A and B. And let's say that A is kind of a sloppy language with a relatively incomplete set of standard libraries, so that it's a pain to use and people when they code they search a certain amount of programming questions that will be answered over the internet on specialized blogs and forums, because of the lack of this information on the language itself or the official documentation (here you can think of PHP or JavaScript for example).

Now let's say that the language B is super neat and clear, easy to understand and use, so that coding with it does not involve searching a lot over the internet. It does involve search for very specialized questions, but overall searching is not the main activity with them as it was for A (here you can think about Golang for example).

Then in that situation, by measuring and comparing volumes of requests, Google Trends for example would always miss the fact that the popularity of each language does not translate to the same amount of search, so there is an extra coefficient that amplifies A when compared to B, hiding the fact that really a lot more people are using B without using search that A, and that difference would be impossible to measure. Let's call that coefficient “sloppiness”.

—

The same difference pops up in different fields as well, for example there is one in politics when polling extremisms. In that situation appears a “shamefulness” coefficient, that pushes the interviewed people to lie when in reality they would vote for extremist parties/candidates. Because 1/ they don't want to admit that they will vote for the “bad guy”, or 2/ they don't want to let anybody know before it happens, if it happens.

In France we had that situation happen for decades with the far right party of “Jean Marie Le Pen” and people hiding in polls their vote. But polls professionals understood that difference because they had a consistent difference between the poll and the vote for decades, so they created a shameful coefficient for this party, and hid that information to the public.

It worked pretty well for decades (polls corrected with this coefficient matching the reality of the vote) but only so long as that party was directed by him, and when his daughter “Marine Le Pen” took the lead it was not shameful to vote for her anymore so what happened is that the people stopped hiding their vote in the polls, and all of them misinterpreted the results of their work and wrongly continued to apply the shameful coefficient, which raised in all the polls the results of the far right higher than it was, higher than it should have ever been.

That created the buzz on all the media, and you know how the media dictate the votes, with or without realizing it, following the rule: the more they talk about something the more people will vote for that, whether they talk about it in good terms or in bad terms it doesn't make a difference (this could be called “the Trump effect”). So that the poor media created the future they feared, and got Trump elected. In France it didn't happen with “Marine Le Pen” but nearly, after claiming for months that the extremists will win, more and more people decided to vote for that.